Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=tgrs20

Download by: [George Mason University] Date: 17 April 2017, At: 08:30

GIScience & Remote Sensing

ISSN: 1548-1603 (Print) 1943-7226 (Online) Journal homepage: http://www.tandfonline.com/loi/tgrs20

Inundation Extent and Flood Frequency MappingUsing LANDSAT Imagery and Digital ElevationModels

Shuhua Qi , Daniel G. Brown , Qing Tian , Luguang Jiang , Tingting Zhao &Kathleen M. Bergen

To cite this article: Shuhua Qi , Daniel G. Brown , Qing Tian , Luguang Jiang , Tingting Zhao &Kathleen M. Bergen (2009) Inundation Extent and Flood Frequency Mapping Using LANDSATImagery and Digital Elevation Models, GIScience & Remote Sensing, 46:1, 101-127

To link to this article: http://dx.doi.org/10.2747/1548-1603.46.1.101

Published online: 15 May 2013.

Submit your article to this journal

Article views: 126

View related articles

Citing articles: 3 View citing articles

101

GIScience & Remote Sensing, 2009, 46, No. 1, p. 101–127. DOI: 10.2747/1548-1603.46.1.101Copyright © 2009 by Bellwether Publishing, Ltd. All rights reserved.

Inundation Extent and Flood Frequency Mapping Using LANDSAT Imagery and Digital Elevation Models

Shuhua QiSchool of Geography and Environment, Jiangxi Normal University, Nanchang, Jiangxi 330027, China

Daniel G. Brown1 and Qing TianSchool of Natural Resources and Environment, University of Michigan,440 Church Street, Ann Arbor, Michigan 48109-1041

Luguang JiangInstitute of Geographic Sciences and Natural Resources Research,Chinese Academy of Sciences, Beijing 100101, China

Tingting ZhaoDepartment of Geography, Florida State University, Tallahassee, Florida 32306-2190

Kathleen M. BergenSchool of Natural Resources and Environment, University of Michigan,440 Church Street, Ann Arbor, Michigan 48109-1041

Abstract: We modeled the extent of inundation around Poyang Lake, China using13 Landsat images and two digital elevation models (DEMs). Boundaries of theobserved inundation extents were (a) labeled with lake-level measurements taken ata representative hydrological station and (b) interpolated to create a Water Line DEM(WL-DEM) that was used to map inundation frequency. A 30 m contour-basedDEM produced slightly better results than the Shuttle Radar Topography MissionDEM, but neither DEM was accurate for medium and low lake levels. The WL-DEMexhibited improved accuracy at medium lake levels, but had relatively high errors atlow lake levels.

INTRODUCTION

Lake floodplains are important environments that: (a) serve as irreplaceable hab-itat for many kinds of waterbirds and other animals (e.g., amphibians), because theirwetlands provide abundant food and shelter; and (b) provide storage for flood waters,

1Corresponding author; email: danbrown@umich.edu

102 QI ET AL.

thereby reducing the magnitudes of floods. These floodplains, often under the controlof built infrastructure (i.e., levees or dams), have been exploited worldwide for foodproduction, reducing these benefits and making agricultural production and associ-ated human settlements vulnerable to flood damage. Knowledge of the spatial andtemporal patterns of flooding, as influenced by the combination of natural floodregimes and human-built controls, is critical in maintaining the ecological functioningof floodplains (Townsend and Walsh, 1998) and predicting the pattern of flood riskfacing the human populations around a lake.

Flood-risk mapping is often the first step for flood-risk management (Plate,2002), and provides needed data for settlement planning, wildlife habitat conservationplanning, design of infrastructure, and flood mitigation and response planning. Riskanalysis yields hazard or risk maps, which can be created using geographic informa-tion systems (GIS) based on hydrological observations of the frequency, magnitude,and duration of flood events, as well as digital elevation models (DEMs) that charac-terize the topographic basin within which flood events occur. Flood-risk maps can beused to identify the weaknesses in a flood defense system and the patterns of vulnera-bility of human infrastructure and wildlife habitat. Because of increasing populationdensities in flood-prone areas throughout the world, including monsoon Asia, as wellas land use and climate changes that affect flood dynamics, and the combination ofthese in historically less information-rich developing countries, efficient food-riskmapping is an increasingly critical tool.

DEMs have been used in various ways to aid flood mapping and modeling. Theyhave been used as integral parts of GIS databases aiming at hydrological flood model-ing (Correia et al., 1998) and for modeling the spatial extent of inundation. However,DEMs are often unavailable or of poor quality (i.e., resolution and accuracy) relativeto needs for modeling, especially in developing countries. Availability has beenimproved with the completion of the Shuttle Radar Topography Mission (SRTM), aninternational research effort that obtained DEMs for nearly the entire globe. SRTMconsisted of a radar system that flew on board the Space Shuttle Endeavour during an11-day mission in February of 2000 (Rabus et al., 2003). SRTM may be used forinundation modeling, but, similar to other DEMs, special attention needs to be paid tothe quality of inundation models based on DEMs in any given situation (Sanders,2007). Because of the characteristic shallow terrain of most floodplains, DEM errors,resolution, or degree of generalization can make it difficult to use DEM-based modelsto successfully describe the detailed geomorphology and flood dynamics for flood-plains (Lee et al., 1992; Hunter and Goodchild, 1995). The waterlines from remotesensing images have also been used to create DEMs (Smith and Bryan, 2007; Kim etal., 2007; Niedermeier et al., 2005). We refer to a DEM created from observed waterelevations as a Water Line DEM (WL-DEM).

Satellite-based remote sensing images have been used to map the extent of floodinundation since the early 1970s (Deutsch et al., 1973; Rango and Solomonson,1974). Most of the early studies used optical remote sensors, such as LANDSAT MSSand TM (Rango et al., 1974; Bhavsar, 1984; Wang et al., 2002). More recent studieshave used radar sensors in place of, or in addition to, optical sensors (Townsend andWalsh, 1998; Sanyal and Lu, 2004; Andreoli et al., 2007). Because optical sensorscannot penetrate clouds, which nearly always accompany flood events, theyhave been mainly used to observe post-flood inundation extent (Marco et al., 2006).

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 103

Optical images also have limited ability to detect water beneath vegetation canopieswith partial or complete vegetation coverage, but they have advantages of relativelyfine spatial resolution (in the case of the series of sensors on Landsat and SPOT) orhigher repeat frequency (in the case of AVHRR and MODIS) that make them usefulin hydrological process studies (Islam and Sado, 2000; Peng et al., 2005). CombiningDEMs with satellite observations of inundation extent offers a promising approach tocharacterizing the spatial and temporal patterns of flood risk (Sanyal and Lu, 2005).

Focusing on Poyang Lake, which is in the Central Yangtze Basin, China, weevaluated the potential of available digital elevation data, in combination with data onhistorical lake-level records and a suite of Landsat TM/ETM+ images, to serve as thebasis for flood-risk analysis. We: (1) extracted the inundated extents from 13 LandsatTM/ETM+ images, obtained at different times with different lake levels, and devel-oped functional relationships between inundated area and lake level for severalhydrological stations to select the gauging station that best represented the dynamicsfor all of Poyang Lake; (2) compared maps of inundated extent derived from twoDEMs, a contour-based DEM and SRTM-DEM, with remote sensing classificationresults to assess their accuracy; (3) created a WL-DEM using remotely sensed inun-dated areas, labeled with the lake-level elevations recorded on the days the imageswere taken, and compared the accuracy of WL-DEM–based inundation to those of theDEM-based inundations; and (4) produced flood frequency maps based on the WL-DEM and on exceedance probabilities of lake levels as measured at the most repre-sentative gauging station. The WL-DEM and flood frequency maps also provide abasis for our future work on bird habitat modeling because the spatial pattern of thevegetation around Poyang Lake is associated with elevation gradients, and differentflood inundation frequencies and durations.

STUDY AREA

The study region for this research is the area around Poyang Lake, which islocated in Jiangxi Province, China (Fig. 1), which we refer to as the Poyang LakeRegion (PLR) in this article. The PLR includes 12 counties and has an area of 20,290km2. Poyang Lake is the largest freshwater lake in China, collects water from five riv-ers (i.e., Ganjiang, Xiushui, Fuhe, Xinjiang, and Raohe), and discharges to the YangtzeRiver along its central reach near the city of Jiujiang. Lake levels at the Duchanggauging station have experienced fluctuations from an average annual low (1955–2000) of 7.6 m above mean sea level (amsl)2 in the winter to an average annual highof 17.3 m amsl in the wetter summer months (Fig. 2), resulting in lake surface-areafluctuations from <1,000 km2 to >4,000 km2 (Shankman and Liang, 2003). There areseveral small water bodies, or sub-lakes, in the basin that are hydrologically con-nected with the main lake when lake levels are high, but become disconnected whenlake levels are low.

Water levels are important determinants of ecological and social processes withinthe PLR. Liu and Ye (2000) found that the spatial pattern of the vegetation aroundPoyang Lake corresponds to elevation gradients and different flood inundation

2All elevations in this paper are reported relative to the National Vertical Datum of China [NVDC],established in 1985.

104 QI ET AL.

frequencies and durations. The migratory waterbirds that over-winter at Poyang Lake,many of which are listed as threatened or endangered, gather in places with similarelevations as they seek water of optimal depth for foraging on underwater vegetation(Liu and Ye, 2000). The aforementioned sub-lakes and large delta floodplains on thewest side of the lake, formed by the sediments of the Ganjiang and Fuhe rivers, areespecially important habitats. Ning et al. (2003) found that the Oncomelania snailsthat play host to schistosome japonium, which is the cause of the disease schistosomi-asis and is endemic in the PLR, are always located on shoals inundated for about 113–216 days a year.

During the past several decades, the hydrological environment in the PLR hasbeen modified greatly. Although land reclamation from the lake began in the Songdynasty (Council of Poyang, 1988; Wu, 1997), many of the current levees aroundPoyang Lake were built during the 1950s and 1960s (Table 1). Under the nationalgovernment’s “Grain First” policy, which was established in the mid-1950s andplaced a high priority on grain production, farmlands were reclaimed from the lake

Fig. 1. Poyang Lake Region (PLR) represented by DEM (lighter shades of grey are higherelevations) and maximum free water surface (in white). Locations of gauging stations areindicated by triangles and river names are indicted in italics.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 105

area and the floodplain was enclosed into many polders that supported high popula-tion densities of about 400–800 persons/km2 (Shankman et al., 2006). It is estimatedthat there were about 565 levees around Poyang Lake in 1998 (Min, 1998), and morethan 330 levees were created during the 1950s and 1960s. As a result of reclamationof land from the lake, the free surface of Poyang Lake was reduced from an estimatedarea of 8,932 km2 to 3,850 km2 (Yu and Dou, 1999; Jiang, 2006). These levees werecharged with protecting about 5,082 km2 of farmland and settlements, and a popula-tion of about 10 million people (Peng, 1999). Most of the empoldered region haselevations in the range from 13 to 16 m.

A second major hydrological change was deforestation, which occurred in thedrainage basin of the lake on a large scale during the 1950s to support the develop-ment of the steel industry. This deforestation resulted in increased runoff and erosionof sediments that were carried to Poyang Lake, and which also caused reduction in itsstorage capacity. As a result of sedimentation, Min (1999) estimated that the storagecapacity of Poyang Lake was reduced by 4.8 percent from 1954 to 1997.

Finally, a number of large hydrological projects have been built in the PoyangLake watershed. The Wan’an reservoir (created in 1993 along the middle reach of the

Fig. 2. Annual maximum, minimum, and average lake levels at the Duchang gauging station.

Table 1. Land Reclamation in the Poyang Lake Region since 1954

Survey year 1954 1957 1961 1965 1967 1976 1984 1995

Poyang Lake area (km2)

5160 5004 4720 4410 4120 3914 3889 3859

Reclaimed area (km2) 156 284 310 290 206 25 30

Total reclaimed area since 1954 (km2)

0 156 440 750 1040 1246 1271 1301

Source: Min, 1998.

106 QI ET AL.

Ganjiang River), and the Zhelin reservoir (created in 1975 at the upper reach of theXiushui River) provide for storage of floodwaters within the basin. Both of these res-ervoirs have reduced sediment delivery to the lake. The Three Gorges Dam is on theYangtze River about 700 km upstream from the outlet of Poyang Lake. One study hasargued that this new dam will intensify floods and droughts in the Poyang Lake flood-plain (Liu and Wu, 1999). Shankman and Liang (2003) argued that, by dischargingwater from behind Three Gorges Dam so that the reservoir has flood storage capacityin July and August (when the upper basin tends to peak), higher water levels in theYangtze River in May and June will meet the traditional flood peaks from the PoyangLake basin and slow Poyang Lake’s discharge to the Yangtze River. Reverse flowfrom the Yangtze River to Poyang Lake is a common phenomenon, but by peakingearlier Yangtze River flow could back up into the lake just as its own discharge is atits highest. Such a scenario would result in increased flood risk around the lake.

These three factors (wetland reclamation for agriculture, deforestation, andhydrologic projects) help explain the increase in annual maximum lake levels during1955–2001 (Fig. 2). During this period, nine extreme (>18.9 m) floods occurred, in1973, 1977, 1980, 1983, 1992, 1995, 1996, 1998, and 1999. The historical high lakelevel occurred during the 1998 flood, when the lake level at Duchang reached 20.6 mand exceeded the previous highest lake level (in 1995) by 0.64 m. It is estimated thatmore than 240 levees in the PLR were damaged in 1998 and more than 722 km2 farm-land was flooded (Jiang, 2006). As the reclaimed land on the Poyang floodplainflooded, more than 4 million households were damaged and more than 1.59 millionpeople were left homeless (UN-Habitat, 2002).

In this study, by creating the WL-DEM and flood frequency maps, we attempt toprovide a basis for our future work on flood risk analysis and bird habitat modelingthat seeks to characterize and assess vulnerability of human and wildlife aroundPoyang Lake. The next section describes the data we used in this study.

DATA

DEM Data

We acquired a DEM for the PLR that had a spatial resolution of 30 m and thathad been interpolated from elevation contours digitized from a map with a scale of1:50,000. The contour interval was 5 m at low elevations and 10 m at higher eleva-tions. We used the DEM both to map inundated extent according to lake level and fororthorectification of remote sensing images. The minimum recorded elevation in theDEM was 12 m, while the lowest observed elevation for Poyang Lake was about 5 m;so the DEM is only suitable for identification of inundated areas when lake levels arehigher than 12 m amsl.

In addition, the SRTM DEM with 3 arc-second (~90 m) spatial resolution wasacquired from USGS EROS Data Center (http://edc.usgs.gov/srtm/data/seam-lesshelp.html). Because random phase noise is a well-known property of SRTM data(Walker et al., 2007), we first conducted phase noise removal by averaging the eleva-tion values in a 3 × 3–pixel window. Overlaying the SRTM data on Landsat TM/ETM+ images and the contour-based DEM revealed a systematic horizontal offsetbetween them. The SRTM data were shifted westward 3 pixels (270 meters) to align

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 107

them with the other data. The vertical datum for the SRTM DEM was the EGM 96geoid. Because we had no direct way to convert from EGM 96 geoid to NVDC 1985,a flat area (15.25 km × 15.25 km) with little vegetation was selected to estimate a con-stant offset for the two vertical datums. We found that the SRTM DEM was 1 meterhigher than the contour-based DEM on average and adjusted the SRTM data by sub-tracting 1 meter. Finally, the SRTM data were resampled to 30 m spatial resolutionusing nearest neighbor resampling so that it could be quantitatively compared withthe contour-based DEM and the Landsat images. The lowest recorded elevation onthe lake in the SRTM DEM was 9 meters. All DEM and image data were reprojectedinto UTM Zone 50N with WGS84 datum.

Levee Data

The levees around Poyang Lake were created to enclose the low-elevation flood-plain for use as farmland, aquaculture, and settlements. These levees, with an averagewidth of about 2~9 m, were classified and mapped (Fig. 3), based on interpretation of

Fig. 3. Polders bounded by different types of levees.

108 QI ET AL.

Landsat TM/ETM+ imagery together with additional information from publishedsources (Jiangxi Province, 1999) and field surveys (Jiang, 2006).

Although the levees around the lake are charged with the protection of more than5,000 km2 of farmland and about 10 million people, most of them are not adequate forthis purpose. Despite being designed for 100-year floods, the current condition of the15 most important levees is such that they can only withstand floods with a recurrenceinterval of 5–8 years (Liu and Xu, 2004). Most of the levees were damaged to someextent, but did not collapse, during flood events in 1992, 1995, 1996, 1998, and 1999(Liu and Xu, 2004).

The PLR levees have been categorized as crucial major levees, major levees, andminor levees according to the amount of enclosed farmland and settlement area.Crucial major levees enclose more than 66.7 km2 farmland, plus big cities or capitalsof counties and are designed to be sufficiently high and hard (i.e., concrete) to with-stand floods with a 100-year recurrence interval. The major levees are also designedto defend against 100-year floods and are charged with protection of more than 33.3km2 farmland. There is less than 33.3 km2 farmland behind the minor levees, andtheir polders are more vulnerable to flooding.

To improve the floodwater storage capacity of the lake, four major levees (Huan-ghu levee, Zhuhu levee, Kangshan dike, and Fangzhouxietang levee) were assigned tofloodwater storage by the government of Jiangxi Province in 1986. According to thepolicy, the storage levees should be breached to allow floodwater to discharge whenlake levels at Hukou (near the lake’s outlet) reach 18.7 m, although they were notbreached during the 1998 flood event because many minor levees and a major leveewere destroyed by the flood (Shankman and Liang, 2003).

After the 1998 flood event, the government carried out a policy of “ReturningFarmland to the Lake,” and many minor polders were abandoned and farmland wasconverted back to wetland. The abandoned levees are classified as “partial return” and“complete return.” In the “partial return” polders, villagers were resettled to higherground, but the farmland can still be cultivated. According to the regulation, when thelake level reaches 18.7 m, partial return levees with protected areas less than 6.67 km2

will be breached. When lake levels reach 19.8 m, the partial return levees with pro-tected areas more than 6.67 km2 will be breached. In the “complete return” polders,the villagers were resettled and the farmland was abandoned to wetland.

Lake-Level Records

Daily lake levels from seven hydrological stations (Duchang, Xingzi, Wucheng[Ganjiang], Wucheng [Xiushui], Tangyin, Kangshan, and Hukou) around PoyangLake (Fig. 1) were obtained from the Hydrological Bureau of Jiangxi Province. TheWucheng station is at the intersection of Ganjiang River and Xiushui River and thereare two gauges that are charged with monitoring these two rivers, respectively. Dailylake-level records were available for 46 years from 1955 to 2001 at each gaugingstation. The published lake-level records for these different stations were basedon different elevation datums and were converted to the NVDC 1985 before furtheranalysis.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 109

METHODS

Image Processing

Thirteen Landsat TM/ETM+ images that cover Poyang Lake (path 121/row 40)were acquired and put to several uses in this study (Table 2). Different subsets ofthese images were used to: (a) select a representative gauging station by relatingremotely sensed inundated area with lake levels; (b) create a WL-DEM and flood fre-quency maps; (c) validate the DEM-based inundated extent models by comparing theDEM-based and remotely sensed inundation extents; and (d) validate the WL-DEM–based inundated-extent model by comparing the WL-DEM–based and remotelysensed inundation extent.

All images were orthorectified using the contour-based DEM. Although LandsatTM near-infrared (NIR) band 4 (0.76–0.90 µm) can usually discriminate water fromland surface because of the low reflectance from water, water can be confused withasphalt in urban areas, which also has low reflectance in NIR. Wang et al. (2002)delineated water successfully by combining TM band 4 and TM band 7 (2.08–2.35µm). However, Poyang Lake at high lake level is an expansive water body that exhib-its enough spectral variation because of differences in sediment concentrations, flooddepths, wetland plants, and discharge from factories and cities to confuse any singleTM band or pair of bands. We delineated inundated extent using unsupervised classi-

Table 2. Dates of TM/ETM+ Images and Usesin the Current Study

Date for images taken Uses in this studya

Dec. 17, 1987 a, bJuly 15, 1989 a, b, cJan. 31, 1993 a, b July 10, 1993 a, b, cApril 4, 1999 a, bNov. 16, 1999 aDec. 10, 1999 a, bApril 16, 2000 a, dJuly 5, 2000 a, b, cAugust 22, 2000 a, b, cSept. 23, 2000 aOct. 9, 2000 a, c, dJan. 29, 2001 a, b

aa = evaluate gauging stations; b = create digital waterbody model and flood-risk map; c = validate the DEM-based inundated extent model; d = validate the DWM-based inundated-extent model.

110 QI ET AL.

fication with all six 30 m spatial resolution bands as input. We used the ISODATAalgorithm and identified 15 spectral classes in the images (Ridler and Calvard, 1978).Because Poyang Lake water exhibits spectral variability, water was included in sev-eral different spectral classes. All spectral classes referring to water, identifiedthrough visual interpretation of a pseudo-color image composite with TM7, TM4, andTM3 were combined to produce binary land-water images for each date (Fig. 4).

Characterizing Lake-Level Dynamics

In order to develop statistical descriptions of lake-level dynamics in the PLR, weselected a single gauging station to represent the dynamics of the entire lake. Loga-rithmic functions were fitted to the relationships between lake levels measured at each

Fig. 4. Remotely sensed inundation extents (in black) from three example images—A. April16, 2000; B. October 9, 2000; C. July 10, 1993—overlaid on a digital elevation model.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 111

station and the remotely sensed inundated areas (km2) using ordinary least-squaresregression, both including and excluding water that was enclosed by levees and,therefore, disconnected from the main lake. The gauging station with the best-fittingfunction was selected as the most representative and used in all subsequent modeling.In order to assess how well this single station represented the lake levels across theentire lake, we measured the differences between lake levels at the selected gaugingstation and those measured at the other stations for all available days.

We used the records from the representative gauging station to calculateexceedance probabilities (EPs). Because the configuration of Poyang Lake was modi-fied by reclamation during the 1950s and 1960s (Table 1), only the lake levelsrecorded during 1970–2001 were used to calculate EPs for all lake levels. The EPs forall lake levels measured on the dates of nine of the TM/ETM+ images (Table 2) werethen assigned to the boundaries derived from those images. We did not use additionalimages because they either represented redundant information in terms of lake levelsor, in two cases, they were set aside for evaluating the accuracy of the WL-DEM. Aflood frequency map was then created using spline interpolation based on the 10 con-tour lines, including the waterline of the modeled maximum inundation extent at a20.6 m lake level in a 1998 flood (ESRI, 2006).

WL-DEM

Because of the 5 m contour interval in the contour-based DEM and the relativelycoarse resolution of the SRTM DEM (90 m), we expected difficulties in characteriz-ing the inundation extents when the lake was at low levels. To improve our ability tomodel medium and low lake levels, we supplemented these DEMs with data from theLandsat images. Assuming that inundation extent is only a function of lake level mea-sured at the representative gauging station, the boundaries of inundation extents fromthe nine TM/ETM+ images used to represent exceedance probabilities (above) weretaken as isolines of lake level. The values of the isolines were assigned as the lakelevel measured at the representative gauging station on the day the image wasobtained. The minimum and maximum lake levels for the imagery dates were 5.69 m(measured at Hukou on January 31, 1993) and 18.43 m (measured at Tangyin on July10, 1993), respectively (Table 3). The WL-DEM was created by spline interpolationusing ArcGIS with the boundaries of remotely sensed inundation extents. To com-plete the WL-DEM, we supplemented it with elevation values from the contour-basedDEM for the regions that were (a) higher than 18.36 m or (b) enclosed by levees.

Evaluating Inundation-Extent Models

Assuming that all areas with elevations lower than the lake level measured at therepresentative gauging station are inundated, we modeled the inundation extent usingthe contour-based DEM, SRTM DEM, and WL-DEM. The Kappa coefficient (Cohen,1960) was used to evaluate the agreement between modeled and remotely sensedinundation extents. The Kappa coefficient reflects the proportion of cells that agree onthe classification as inundated or not, after chance agreement is removed. We refer tothe general benchmarks for interpreting Kappa suggested by Landis and Koch (1977),where 0.41–0.60, 0.61–0.80, and 0.81–1.00 can be interpreted as indicating moderate

112 QI ET AL.

Tabl

e 3.

Lak

e Le

vels

Mea

sure

d at

Sev

en H

ydro

logi

cal S

tatio

ns a

nd R

emot

ely

Sens

ed In

unda

tion

Are

as fo

r eac

h La

ndsa

t Im

age

Dat

e

Lake

leve

l, in

m:

Dat

eX

ingz

iW

uche

ng

(Xiu

shui

)W

uche

ng

(Gan

jiang

)Ta

ngyi

nK

angs

han

Duc

hang

Huk

ouIn

unda

ted

area

(k

m2 )a

Inun

date

d ar

ea

(km

2 )b

Dec

. 17,

198

78.

2810

.04

10.1

610

.90

12.0

59.

737.

5114

89.4

2613

.9Ju

ly 1

5, 1

989

17.3

617

.35

17.3

517

.42

17.2

717

.35

17.5

633

30.9

4925

.8Ja

n. 3

1, 1

993

6.65

9.08

9.21

10.0

111

.11

8.32

5.69

986.

618

66.2

July

10,

199

318

.29

18.3

218

.33

18.4

318

.32

18.3

618

.26

3522

.465

30.4

Apr

il 6,

199

910

.18

11.4

311

.55

12.3

413

.13

11.4

87.

3520

81.7

3354

.7N

ov. 1

6, 1

999

11.0

211

.38

11.3

911

.53

12.0

211

.27

10.9

619

33.2

3250

.4D

ec. 1

0, 1

999

8.72

9.81

9.74

10.2

911

.46

9.29

8.61

1312

.926

42.9

Apr

il 16

, 200

010

.45

11.7

710

.72

11.9

712

.95

11.2

39.

9220

90.5

3080

.5Ju

ly 5

, 200

015

.54

15.5

215

.55

15.5

915

.45

15.5

415

.68

3152

.442

81.0

Aug

. 22,

200

013

.73

13.8

313

.81

13.8

213

.72

13.7

213

.85

2668

.241

59.7

Sept

. 23,

200

014

.13

14.1

714

.16

14.2

514

.11

14.1

614

.23

3197

.347

88.0

Oct

. 9, 2

000

14.1

414

.15

14.1

313

.22

13.2

414

.13

14.2

830

01.1

4336

.9Ja

n. 2

9, 2

001

8.72

11.3

511

.43

12.0

413

.09

11.0

38.

4921

11.6

3957

.1

a Excl

udin

g ar

ea e

nclo

sed

by le

vee.

b Incl

udin

g ar

ea e

nclo

sed

by le

vee.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 113

agreement, substantial agreement, and almost perfect agreement, respectively.Because we are performing many comparisons with varying population sizes (i.e.,number of pixels; described below), we did not attempt to perform significance tests,but rather use Kappa as an indicator of relative strength of agreement among the dif-ferent models.

A complicating factor in this analysis is that the accuracies are spatially heteroge-neous, i.e., regions far away from the boundary of the water body will have higheraccuracies than regions near the boundary. To characterize this heterogeneity and bet-ter understand the strength of agreement, buffers with distances of 1,2,…,10 pixelswere created from the remotely sensed water boundary and the Kappa coefficient wasused to evaluate the accuracy of the modeled water body in each of these buffers (i.e.,all pixels within the specified distance were included in the Kappa calculation). Wereport Kappa as a function of buffer distance from the observed water boundary.Because the pixels nearest to the water boundary are those most likely to be misclassi-fied by the model, the error rate should increase (and the accuracy decrease) as theanalysis focuses on those pixels (i.e., as the buffer distance decreases). For flood-riskmapping in land use management and emergency response, measuring accuracy overthe narrowest distance buffers imposes a standard that is too high; rather understand-ing the general patterns of inundation (e.g., averaged over larger areas) is often suffi-cient. Therefore, considering the levels of agreement at larger distances should besufficient. However, if flood-risk mapping is to be used for engineering purposes,accuracy standards should be higher and accuracy statistics should be interpreted atthe narrowest distance bands. In any case, including all pixels in the region in theanalysis (i.e., not using buffers to focus the analysis) imposes a standard that is toolow, because all areas that are correctly identified as not inundated because they arenowhere near the water boundary are included in the analysis.

We compared the inundation extent modeled with each of the three DEMs withthat observed on a number of Landsat images. We selected five images to represent arange from medium to high lake levels to test the DEM-based model (Table 2):August 22, 2000; October 9, 2000; July 5, 2000; July 15, 1989; and July 10, 1993.Because the water bodies enclosed by levees have no direct hydrological connectionwith Poyang Lake, we did not attempt to model inundation area behind the levees onthe basis of lake levels and therefore compared inundation extents at places along thelake boundary with no levees present. The result is a conservative estimate of theaccuracy of the inundation extent of the main lake, however, because inundation ofareas bounded by levees can be mapped with a high degree of accuracy given infor-mation about the location and height of levees. We did not try to model the inundationextent for low lake levels because the contour-based and SRTM DEM data limit ourability to represent only those areas above 12 m and 9 m, respectively. The WL-DEMwas created to improve our ability to model inundation areas at lower water levels,and was based on inclusion of water boundary contours from 10 images. We com-pared WL-DEM–based inundation extents with extents observed in two scenes thatwere not used in the creation of the WL-DEM and that represented medium and lowlake levels.

114 QI ET AL.

RESULTS AND DISCUSSION

Representative Gauging Station

The fit of the relationship between lake levels and inundated areas as measuredfrom Landsat images (Fig. 5) was used as a measure of how representative each ofseven hydrological stations was for modeling the entire lake (Table 3). The increase ininundated area with lake level was logarithmic in all cases, and the relationshipsbased on inundated area that excluded inundated areas enclosed by levees fit betterthan those that included enclosed water for all but the Kangshan gauge. The waterbodies behind levees are mainly natural or human-made ponds, and these water bod-ies have no direct hydrologic connection with Poyang Lake.

Among the seven gauging stations, the lake level from Duchang was best corre-lated with inundated area (R2 = 0.965). Thus, we concluded that Duchang is the mostrepresentative station that can be as input to DEM-based models of inundated areawithin the area outside the levees. This result is in agreement with those of Min(2004). Duchang is located near the middle of the eastern lakeshore (Fig. 1). The lakelevels from Duchang were used to drive subsequent inundation-extent models andflood frequency maps.

The median differences and variation in differences between Duchang lake levelsand measurements made at the other six gauges during 1955–2001, which reflect thedegree to which the Duchang measurements inaccurately represent the entire lake sur-face, decreased with increasing lake level (Fig. 6). The maximum difference wasabout 5 m at the lowest measured lake levels (~6 m). These results indicate that theDuchang station represents well the levels across the entire lake at levels above about11 m, but that at levels of 11 m or less, there is substantial heterogeneity in the lakelevels. This limits the ability of models based on lake levels measured at a singlegauging station to appropriately represent the entire lake at low lake levels, but sug-gests that such models can be reasonably useful for representing lake dynamics athigher lake levels.

DEM-Based Models of Flood-Inundation Extent

The Kappa coefficients resulting from the comparison of contour-based (Fig. 7A)or SRTM-based (Fig. 7B) DEM inundation extents with the satellite classificationsfor the five test dates always increased as pixels at increasing distances from theobserved water boundary were included in the calculation. This suggests that the suit-ability of inundation extent models for any given purpose will depend on the amountof spatial specificity required from the model. In all cases, the modeled inundatedareas exhibited a substantial level of agreement (Kappa > 0.6) with image data, whenobserved over a 600 m band around the observed water boundary (i.e., 10 pixels oneach side of the boundary). However, when observed over a 60 m band, there was noteven moderate agreement in most cases. This indicates a non-homogeneous surface ofthe lake, i.e., the lake water level is dynamic and complicated.

For the models based on the DEMs, the Kappa coefficients corresponding todates with higher lake levels (i.e., July 10, 1993 and July 15, 1989) were always

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 115

higher than those for dates with lower lake levels (i.e., August 22, 2000 and October9, 2000) at the same buffer distance. Although all dates included in this analysis hadlake levels above 11 m, the general trend of increasing spatial heterogeneity of lake

Fig. 5. Fitted relationships between remotely sensed inundation extents and lake levels forseven hydro-stations in Poyang Lake. Open squares represent data that include water enclosedby levees and solid diamonds exclude the water enclosed by levees.

116 QI ET AL.

levels at lower levels (Fig. 4) might still cause a violation of the assumption of con-stant lake levels and, therefore, explain the lower levels of agreement at medium andlow lake levels. The increasing accuracy of the DEMs in representing inundationareas at high lake levels suggests that there is some promise in using such models forflood-risk mapping where the concern is with mapping risks associated with the high-est lake levels, although the DEMs alone do not appear to provide sufficient detail formapping inundation areas at low and medium lake levels.

The modeled inundation extents based on the SRTM DEM were always lowerthan those using the contour-based DEM, but the differences generally declined asmore pixels were included in the calculation (i.e., at larger buffer distances) and atlower lake levels. Accuracies of inundated areas based on the SRTM DEM that wereaveraged over a distance of two or more pixels from the water boundary were roughlycomparable to those from the contour-based DEM (Fig. 7). These results suggest thatthe SRTM-based DEM is not as good as the contour-DEM in providing a model ofinundated area, but that the differences in accuracy were relatively small. This is atleast partially related to differences in resolution, and it could be that the 30 m SRTM-based DEM, which was not available for our study, could be as good as or better thanthe contour-based DEM. Further analysis of the accuracy of the SRTM DEM formodeling flooding extent will be important for extending the type of analysis pre-sented here to the broader geographic extent covered by the SRTM data.

Slope angle is likely to have some effect on the accuracy of inundation modelingat low lake levels based on DEM. As slope decreases, vertical error would naturallyeffect the horizontal position of the flood extent at an increasing rate (Hodgson andBresnahan, 2004). However, we observed that the slope at the water boundary of

Fig. 6. Box plot of differences between measurements made at Duchang and the six othergauging stations for all days between 1955 and 2001.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 117

flood extent is quite homogeneous. Pixels in the 10-pixel buffer had an average slopeof 0.75°, with a standard difference of 0.021. Because of this relative homogeneity,variability in the slope in the Poyang Lake Region had a negligible effect on variabil-ity in the accuracy of the flooding extent models.

WL-DEM–Based Inundation Extent Modeling

Because the WL-DEM (Fig. 8) includes DEM elevations at higher elevations(above 18.36 m), the inundation extents based on the WL-DEM should be similar tothose based on the DEM in high-water conditions, and we evaluated modeled inunda-tion extents from the WL-DEM for only medium and low lake levels. Also, because

Fig. 7. Kappa coefficients for DEM-modeled inundation extents compared to Landsat imagesfrom five different dates. A. Contour-based DEM. B. SRTM. Distance buffers were definedrelative to the remotely sensed water-body boundary.

118 QI ET AL.

the contour-based DEM produced better inundation estimates than the SRTM-DEM,only the inundation extents from the contour-based DEM were compared with WL-DEM–based inundation extents. Comparing the DEM-based and WL-DEM–basedmodeled inundation extents for one date (Fig. 9), it is clear that the WL-DEM–basedinundation extent exhibits much more spatial detail near the mouths of the Ganjiangand Fuhe rivers than the DEM-based model. The comparison of Kappa coefficientsfor the contour-DEM–based and WL-DEM–based inundation extents also shows thatthe WL-DEM–based model is slightly better than the DEM-based model at mediumlake levels (Fig. 10). The WL-DEM–based model had a moderate level of agreementwith the remotely sensed inundation extent taken at a medium lake level, at distancesgreater than 60 m (i.e., 2 pixels) from the water boundary.

Because of the apparently low level of accuracy of the DEM-based inundationmodels at low lake levels, we did not compare the DEM-based and WL-DEM-basedmodels for low lake levels. The WL-DEM–based inundation extent had a reasonable

Fig. 8. The Water Line DEM for PLR.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 119

amount of detail for the water body at the mouth of the river at a relatively low lakelevel (Fig. 11). The Kappa coefficient suggests that the WL-DEM was not able toaccurately represent inundation extents at low lake levels (Fig. 12). The resultinginundation extent map had moderate levels of agreement with the remotely sensed

Fig. 9: The DEM-based (A) and DWM-based (B) modeled inundation extent for October 9,2000 (Duchang level = 14.13 m).

Fig. 10. Kappa coefficients for DEM-based and DWM-based inundation extents forOctober 9, 2000 within buffer distances from the remotely sensed water-body boundary.

120 QI ET AL.

data at distances greater than 120 m (i.e., four pixels) from the water boundary andreached a Kappa of 0.6 only at a buffer distance of 300 m (i.e., 10 pixels).

The presence of levees, along with the complex hydrology of the lake, compli-cates our ability to represent its dynamics with simple inundation models of this sort.Our analysis the agreement of inundation areas focused on parts of the lake boundarywithout levees present. A complete model of inundation must consider the constraintson flow imposed by the levee system. Furthermore, several smaller lakes around theboundary of Poyang Lake, such as Bang Lake, are hydrologically disconnected fromthe main lake. Hu et al (1997) has compared the lake level from Bang Lake and theWucheng (Xiushui) hydrological station and found that the Bang Lake level does nottrack with the Wucheng (Xiushui) lake level when lake levels are low. For that reasonthere is a weak relationship between the Duchang lake level and the inundation areaof the sub-lakes. For example, the water body of Bang Lake (Fig. 13) was wider onJanuary 29, 2001 (lake level at Duchang was 11.03 m) than on April 6, 1999, whenthe Duchang Lake’s level was higher (11.48 m). This is an important source of errorin the WL-DEM–based inundation extent model for the sub-lakes, and limits its use-fulness for modeling bird habitats around these sub-lakes.

Fig 11. DWM-based modeled inundation extent for April 16, 2000 (Duchang level = 11.23 m).

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 121

Exceedance Probabilities and Flood Frequency Mapping

We calculated exceedance probabilities based on the daily lake-level records atDuchang from 1970 to 2001 and, because flooding events happen in June, July andAugust, separately for those months (Fig. 14). Taking the boundaries of remotelysensed inundation extents as EP isolines (based on Table 4), we mapped flood fre-quencies for annual, June, July, and August conditions (Fig. 15) within the boundaryof modeled inundation extent for historic maximum lake level measured at Duchang

Fig. 12. Kappa coefficients for DWM-based inundation extents for April 16, 2000 (Duch-ang level = 11.23 m) within buffer distances from remotely sensed water-body boundary.

Fig. 13. An example of the inundation extent for a sub-lake (Bang Lake) of Poyang Lake.

122 QI ET AL.

(20.60 m). Because the lowest lake level of the permanent water body (6.28 m),which occurred on December 29, 1979, is below the minimum elevation in the DEMand because no remote sensing images were available on December 29, 1979, wewere not able to distinguish the permanent water area with EP = 1 from the yearlyflood frequency with EP = 0.919. This approach is useful, however, for the purposesof evaluating flood risk, because the minimum lake levels in June, July, and Augustare all greater than 8.32 m, so the monthly permanent water body can be mapped forthe flood season.

Fig. 14. The exceedance probabilities for every Duchang Lake level from 1970–2001 records,annually and for June, July, and August.

Table 4. Exceedance Probabilities for Lake Levels Associated with Landsat Image Dates

Image datesDuchang lake

level, m EP yearly EP in June EP in July EP in August

Dec. 17, 1987 9.73 0.774 1 1 1July 15, 1989 17.35 0.045 0.054 0.274 0.021Jan. 31, 1993 8.32 0.919 1 1 1July 10, 1993 18.36 0.021 0.020 0.144 0.054April 6, 1999 11.48 0.562 0.966 0.998 0.962Dec. 10, 1999 9.29 0.816 1 1 1July 5, 2000 15.54 0.137 0.260 0.605 0.398Aug. 22, 2000 13.72 0.294 0.585 0.890 0.702Jan. 29, 2001 11.03 0.623 0.988 1 0.972

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 123

CONCLUSIONS

Thirteen Landsat TM/ETM+ images were used to identify the inundated area inand around Poyang Lake. These areas were regressed against measured lake levelsfrom seven hydrological stations in Poyang Lake. From the correlation coefficients,we concluded that the Duchang station was the most representative hydrological sta-tion for evaluating inundation extent.

Based on lake levels measured at Duchang, we used two approaches for mappingthe inundation extent for Poyang Lake floodplain and compared their results with theremotely sensed inundation extents as reference. Because the lake levels are more

Fig. 15. The Annual (A), June (B), July (C), and August (D) inundation probability for thePoyang Lake region.

124 QI ET AL.

variable as lake level decreases, the inundation extent from the DEM-based model ismost reasonable for high lake level conditions, but is not suitable for representinginundation extent at medium and low lake levels. The results of inundation modelingusing the SRTM DEM were not as good as the contour-based DEM for the floodplainarea, but the SRTM DEM may be a reasonable substitute where other sources areunavailable. Because we did not have access to the 30 m resolution SRTM DEM, wecan only speculate that some of the difference in accuracy is due to the coarser resolu-tion of the SRTM DEM.

Using the boundaries of remotely sensed inundation extents as isolines of lakelevel, a digital water-level model (WL-DEM) was created for modeling inundationextent at medium and low lake levels. The WL-DEM–based inundation extents werebetter than the DEM-based models at both low and medium lake levels. These resultspoint to an important advantage of combining digital elevation data with remotelysensed images of inundation in representing flood dynamics, and suggest the possibil-ity of using such data to improve habitat characterizations at low lake levels. Informa-tion on lake levels, combined with the WL-DEM and a map of levees with heights,can therefore do a reasonably good job of representing flood inundation aroundPoyang Lake when lake levels are high and where levees are not present (i.e., wherewe evaluated accuracy). This is useful because flood-risk assessments are generallyconcerned with high, rather than low, lake levels. When lake levels are low, the inun-dation extent in the sub-lakes cannot be improved by using WL-DEM, because of thehydrological disconnection between the sub-lakes and the main lake at low lake lev-els. The WL-DEM provides additional details in the main lake at low lake levels andincreases the inundation-extent model accuracy at medium lake levels. Because thehabitats for most waterbirds are below 18 meters (Liu, 2000), the WL-DEM is usefulfor bird habitat modeling. However, habitat for wetland birds that over-winter in theregion when lake levels are low are very sensitive to variations in lake level. Futurework should build on more extensive archives of satellite data from multiple plat-forms to build more complete representations of the flood dynamics in the PoyangLake Region (Andreoli et al., 2007) and elsewhere. To help bird habitat modelingaround sub-lakes, separate hydrological gauging stations are required to accuratelyrepresent lake levels and inundation areas within these sub-lakes.

The annual and flood-season monthly flood frequencies were mapped by usingthe remotely sensed inundation extents as isolines and assigning them with lake-levelexceedance probability estimated from Duchang data. These flood frequency mapswill be used as important inputs to integrated assessment of flood dynamics, ecologi-cal processes, and human vulnerabilities, and can be used in planning, design, andmanagement of flood works, natural reserves, and land use and land cover policies.

ACKNOWLEDGMENTS

We acknowledge financial support for this project from the U.S. National Aero-nautics and Space Administration (NASA; Grant No. NEWS/04-1-0000-0025), theU.S. National Science Foundation (NSF; Grant No. BCS-0527318), and the NationalScience Foundation of China (NSFC; Grant No. 40561011). We also wish to thankShuming Bao, Lin Zheng, and Ying Liu for their collaboration and for arranginglogistical support in the field.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 125

REFERENCES

Andreoli, R., Yesou, H., Li, J., and Y. Desnos, 2007, “Synergy of Low and MediumResolution ENVISAT ASAR and Optical Data for Lake Watershed Monitoring:Case Study of Poyang Lake (Jiangxi, P.R. China),” paper presented at ENVISATSymposium 2007, Montreaux, Switzerland, April 2007.

Bhavsar, P., 1984, “Review of Remote Sensing Applications in Hydrology and WaterSources Management in India,” Advances in Space Research, 4(11):193–200.

Cohen, J, 1960, “A Coefficient of Agreement of Nominal Scales,” Educational andPsychological Measurement, 20:37–46.

Council of Poyang Lake Research Editors, 1988, Researches on Poyang Lake, Shang-hai, China: Shanghai Science & Technology Press, 118 p. (in Chinese).

Correia, F. N., Rego, F. C., Saraiva, M. D. S., and I. Ramos, 1998, “Coupling GISwith Hydrologic and Hydraulic Flood Modeling,” Water Resources Management,12:229–249.

Deutsch, M., Ruggles, F., Guss, P., and E. Yost, 1973, “Mapping the 1973 MississippiFloods from the Earth Resource Technology Satellites,” in: Proceedings of theInternational Symposium on Remote Rensing and Water Resource Management,Burlington, Canada: American Water Resource Association, No. 17, 39–55.

ESRI, 2006, ArcGIS v. 9.2., Redlands, CA: Environmental Systems Research Insti-tute.

Gianinetto, M., Villa, P., and G. Lechi, 2006, “Postflood Damage Evaluation UsingLandsat TM and ETM+ Data Integrated with DEM,” IEEE Transactions on Geo-science and Remote Sensing, 44(1):236–243.

Hodgson, M. E. and P. Bresnahan, 2004, “Accuracy of Airborne Lidar Derived Eleva-tion: Empirical Assessment and Error Budget,” Photogrammetric Engineeringand Remote Sensing, 70(3):331–339.

Hu, C. H., Jiang, J. H., and H. H. Zhu, 1997, “Analysis on Water-Level Relationshipsbetween Banghu Depression and Poyang Lake and Its Submission and Emersionof Bottomland,” Oceanology et Limnology Sinica, 28(6):617–623 (in Chinese).

Hunter, G. J. and M. F. Goodchild, 1995, “Dealing with Error in Spatial Databases: ASimple Case Study,” Photogrammetric Engineering and Remote Sensing,61(5):529–537.

Islam, M. D. M., and K. Sado, 2000, “Development of Flood Hazard Maps of Bangla-desh Using NOAA-AVHRR Images with GIS,” Hydrological Sciences Journal,45(3):337–355.

Jiang, L. 2006, “Flood Risk and Land Use Change in the Wetland Restoration AreaAround Poyang Lake, China,” Ph.D. dissertation, Institute of GeographicalScience and Natural Resources Research, Chinese Academy of Sciences,Beijing, China (in Chinese).

Jiangxi Province Department of Water Resources, 1999, Levee Atlas of Jiangxi Prov-ince, Nanchang, China: Jiangxi Province Department of Water Resources (inChinese).

Kim, D. J., Moon, W. M., Park, S. E. Kim, J.-E., and H. S. Lee, 2007, “Dependence ofWaterline Mapping on Radar Frequency Used for SAR Images in IntertidalAreas,” IEEE Geoscience and Remote Sensing Letters, 4(2):269–273.

126 QI ET AL.

Landis, J. R. and G. G. Koch, 1977, “The Measurement of Observed Agreement forCategorical Data,” Biometrics, 33:159–173.

Lee, J., Snyder P. K., and P. F. Fisher, 1992, “Modeling the Effect of Data Errors onFeature Extraction from Digital Elevation Models,” Photogrammetric Engineer-ing and Remote Sensing, 58:1461–1467.

Liu, R. and C. Xu, 2004, “The Anti-2 Seepage Design and Analysis of the LeveeBody of Liuhu Dike on the Second Stage Flood Control Project of Poyang Lake,”Jiangxi Hydraulic Science & Technology, 30: 54–60 (in Chinese).

Liu, X. and D. Wu, 1999, “Brief Analysis of the Effect of the Poyang Lake WaterLevel in Flood Period by the Three-Gorges Project,” Jiangxi Hydraulic Scienceand Technology, 25(2):71–75 (in Chinese).

Liu, X. Z. and J. X. Ye, 2000, The Jiangxi Wetland, Beijing, China: China Forest Pub-lishing Company (in Chinese).

Min, Q., 1998, “Analysis on the Influence of Reclamation on Flood in Poyang LakeRegion,” Jiangxi Hydraulic Science & Technology, 24(3):177–185 (in Chinese).

Min, Q., 1999, “Evaluation of the Effects of Poyang Lake Reclamation on Floods,”Yangtze River, 30(7): 30–32 (in Chinese).

Min, Q., 2004, “Analysis and Calculation of Frequency Variation of Flood Stages inPoyang Lake,” Hydrology, 24(4): 17–20.

Niedermeier, A., Hoja, D., and S. Lehner, 2005, “Topography and Morphodynamicsin the German Bight Using SAR and Optical Remote Sensing Data,” OceanDynamics, 55(2):100–109.

Ning, A., Chen, N. G., Zhong, J. H., Chen, H. G., Lin, D. D., Liu, H. Y., Gao, Z. L.,Liu, Y. M., Hu, F., and S. J. Zhang, 2003, “Study of the Relationship BetweenSnail Distribution of Marshland and Changes of Water Level in Poyang LakeRegion,” China Journal of Schistomiasis Control, 15(6):429–433 (in Chinese).

Peng, D. Z., Xiong, L. H., Guo, S. L., and N. Shu, 2005, “Study of Dongting LakeArea Variation and Its Influence on Water Level Using MODIS Data,” Hydrolog-ical Sciences Journal, 50(1):31–44.

Plate, E. J., 2002, “Flood Risk and Flood Management,” Journal of Hydrology,267:2–11.

Rabus, B., Eineder, M., Roth, A., and R. Bamler, 2003, “The Shuttle Radar Topogra-phy Mission—A New Class of Digital Elevation Models Acquired by Space-borne Radar,” ISPRS Journal of Photogrammetry and Remote Sensing,57(4):241–262.

Rango, A. and V. V. Solomonson, 1974, “Regional Flood Mapping from Space,”Water Resource Research, 10(3):473–484.

Ridler, T. W. and S. Calvard, 1978, “Picture Thresholding Using an Interactive Selec-tion Method,” IEEE Transactions on Systems, Man, and Cybernetics, Smc-8(8):1264–1291.

Sanders, B. F., 2007, “Evaluation of On-line Dems for Flood Inundation Modeling,”Advances in Water Resources, 30(8):1831–1843.

Sanyal, J., and X. X. Lu, 2005, “Remote Sensing and GIS-Based Flood VulnerabilityAssessment of Human Settlements: A Case Study of Gangetic West Bengal,India,” Hydrological Processes, 19(18): 3699–3716.

INUNDATION EXTENT AND FLOOD FREQUENCY MAPPING 127

Sanyal, J., and X. X. Lu, 2004, “Application of Remote Sensing in Flood Manage-ment with Special Reference to Monsoon Asia: A Review,” Natural Hazards,33(2):283–301.

Shankman, D., Keim, B. D., and J. Song, 2006, “Flood Frequency in China’s PoyangLake Region: Trends and Teleconnections,” International Journal of Climatol-ogy, 26:1255–1266.

Shankman, D. and Q. Liang, 2003, “Landscape Changes and Increasing Flood Fre-quency in China’s Poyang Lake Region,” Professional Geographer, 55(4):434–445.

Smith, R. K. and K. R. Bryan, 2007, “Monitoring Beach Face Volume with a Combi-nation of Intermittent Profiling and Video Imagery,” Journal of CoastalResearch, 23(4):892–898.

Townsend P. A. and S. J. Walsh, 1998, “Modeling Floodplain Inundation Using anIntegrated GIS with Radar and Optical Remote Sensing,” Geomorphology,21:295–312.

UN-Habitat, 2002, “Best Practice—Post-flood Resettlement in the Poyang LakeRegion,” Jiangxi Province Together Foundation and UN-Habitat [http://database.bestpractices.org/bp_display_best_practice.php?best_practice_id=61].

Wang, Y., Colby, J. D., and K. A. Mulcahy, 2002, “An Efficient Method for MappingFlood Extent in a Coastal Floodplain Using Landsat TM and DEM Data,” Inter-national Journal of Remote Sensing, 23(18):3681–3696.

Walker, W. S., Kellndorfer, J. M.., and L. E. Pierce, 2007, “Quality Assessment ofSRTM C- and X-band Interferometric Data: Implications for the Retrieval ofVegetation Canopy Height,” Remote Sensing of Environment, 106(4):428–448.

Wu, J., 1997, “Economic Status in Poyang Lake Region at Tang and Song Dynasty,”Journal of Shanghai Normal University, 1:43–50 (in Chinese).

Yu, X. and H. Dou, H., 1999, “Impact of Reclamation of Poyang Lake on Flood andCountermeasures,” Environment Protection, 10:6–7 (in Chinese).